pi-elevenlabs-tts

Pi extension that reads assistant text responses aloud with ElevenLabs text-to-speech. Tool calls and tool results are ignored. Bare http:// and https:// URL schemes are removed before speech. Current playback is stopped when the user submits a new message. On macOS, TTS skips playback by default when pmset reports active system audio or WebRTC audio activity, and the Pi status line shows when audio is blocking speech.

Requirements

• Pi (@mariozechner/pi-coding-agent)
• ElevenLabs API key
• An audio player:
  • macOS: built-in afplay
  • Linux/other: mpv or ffplay, or set ELEVENLABS_PLAYER

Setup

Install and authenticate the ElevenLabs CLI. The CLI is useful for login/account setup; TTS behavior is configured in the Pi extension via environment variables and /speak commands.

npm install -g @elevenlabs/cli
elevenlabs auth login

The extension reads the API key from either:

1. ELEVENLABS_API_KEY, or
2. ~/.elevenlabs/api_key created by elevenlabs auth login

Try locally

From this package directory:

pi -e ./extensions/elevenlabs-tts/index.ts

Or install as a local Pi package:

pi install /Users/ajsharp/code/pi-elevenlabs-tts

Then restart Pi or run /reload.

Commands

Inside Pi:

/speak status                  # show current config
/speak on                      # enable auto-read
/speak off                     # disable auto-read
/speak stop                    # stop playback
/speak last                    # replay last spoken assistant text
/speak voices [search]         # list available ElevenLabs voices
/speak voice <voice-id>        # set voice
/speak model <model-id>        # set TTS model
/speak skip-when-audio on|off  # skip TTS when macOS reports active audio
/speak stability <0-1>         # set voice stability
/speak similarity <0-1>        # set similarity boost
/speak style <0-1>             # set style exaggeration
/speak speed <0.7-1.2>         # set speaking speed
/speak boost on|off            # speaker boost
/speak reset                   # reset to environment/default config

Command changes are persisted into the current Pi session and restored after /reload.

Configuration

Environment variables:

Variable	Default	Description
ELEVENLABS_API_KEY	read from ~/.elevenlabs/api_key	ElevenLabs API key
ELEVENLABS_TTS_ENABLED	true	Auto-read responses aloud
ELEVENLABS_VOICE_ID	JBFqnCBsd6RMkjVDRZzb	Voice ID
ELEVENLABS_MODEL_ID	eleven_turbo_v2_5	TTS model
ELEVENLABS_OUTPUT_FORMAT	mp3_44100_128	ElevenLabs output format
ELEVENLABS_API_BASE	https://api.elevenlabs.io	API base URL
ELEVENLABS_MAX_CHARS	4000	Max characters spoken per response
ELEVENLABS_STRIP_CODE_BLOCKS	true	Remove fenced code blocks from spoken text
ELEVENLABS_SKIP_WHEN_AUDIO_ACTIVE	true	On macOS, skip TTS when pmset reports active output audio or WebRTC audio activity
ELEVENLABS_AUDIO_ACTIVE_STATUS_INTERVAL_MS	30000	Status-line polling interval for macOS audio activity
ELEVENLABS_STABILITY	ElevenLabs default	Voice stability, 0-1
ELEVENLABS_SIMILARITY_BOOST	ElevenLabs default	Similarity boost, 0-1
ELEVENLABS_STYLE	ElevenLabs default	Style exaggeration, 0-1
ELEVENLABS_SPEED	ElevenLabs default	Speaking speed, usually 0.7-1.2
ELEVENLABS_USE_SPEAKER_BOOST	ElevenLabs default	true/false speaker boost
ELEVENLABS_PLAYER	afplay on macOS, mpv --really-quiet elsewhere	Audio player command

Example:

export ELEVENLABS_VOICE_ID="JBFqnCBsd6RMkjVDRZzb"
export ELEVENLABS_MODEL_ID="eleven_turbo_v2_5"
pi -e ./extensions/elevenlabs-tts/index.ts